Hi Mike

The idea that intellgences are created to operate in classes of environments seems reasonable.  But the environments that exist in a place as localised and moderately uniform as the Earth are very diverse.  Which is why we have plus or minus about 1--20 million species of life on this planet.  Needless to say, the minds of these species are quite diverse.

> An artificial mind that has been programmed with all of our knowledge
> about our environment and all of our skills at problem solving our
> problems will be like us except for the advantages supplied by the
> machine hardware.  These advantages are limited in number and very
> specifically describable and predictable and understandable.  There is
> ONE general organizational structure that optimizes this AGI for our
> environment.  All deviations from the one design only serve to make
> the AGI function less effectively.

This argument seems to me to be very similar to the one used by many mainstream economists that there is only one optimally functioning economy (free market based etc. etc.) and that any deviation from the idea will reduce the productivity of the economy.

This is based on a very simplistic interpretation of optimisation theory - that in effect says that if you knew everything and could do anything you wanted to do, you would find only one best optimisation state for any set of goals.  But this is about as useful as saying that we should start by assuming we are all God.

We live in a reality in which we cannot know everything and in which we cannot do everything that we want to - and this will continue to apply to super intelligent AGs that exceed human capacity (because the introduction of super AGIs imediately makes the environment more complex than it was before).  This notion that we live in a world where perfect optimisation is never possible is called in economics the "theory of second best".  But this name is actually a bit misleading.  The inevitable limitations on what we can know and what we can do are so great that we are nowhere near finding second best solutions - we should think notionally in terms of nth best.

What this means then is that it is quite possible that clever thinking can come up with all sorts of quite different but in a sense equally useful solutions to any optimisation issues - especially if we are dealing with very compex issues - ie. the environments are complex and the goals are complex.

So my guess is that there is a high probablility that the intelligent minds that emerge from AGI work will be very diverse in their characteristics and behaviour and that many will not be much like us.

Apart from anything else, an environment into which super intelligent AGIs are introduced is no longer the environment for which human intelligence was optimised by the slow and constrained process of evolution.

Cheers, Philip

Reply via email to